The Power of Ensembles for Active Learning in Image Classification

By William H. Beluch et al

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Related Work
3. Methodology
4. Experimental Results

Summary

Deep learning methods have become the de-facto standard for challenging image processing tasks such as image classification. Active learning techniques can alleviate the labeling effort required for large datasets. This paper investigates ensemble-based methods for active learning in image classification using convolutional neural network classifiers. The study compares ensemble methods with Monte-Carlo Dropout and geometric approaches, finding that ensembles perform better. Results on MNIST, CIFAR-10, and ImageNet datasets show promising outcomes with ensemble-based active learning. The paper also discusses the challenges and strategies for active learning in the context of deep neural networks and high-dimensional data.